Experimental Short-Term Climate Forecasting Procedure for US Winter Temperatures and Its Verification

摘要

A linear multivariate regression model for continental United States Januray surface air temperatures is presented. The predictors enter the regression equation at more than one lag if their effects on the predictand tend to be spread over several months rather than be completed in one month. The monthly mean surface tempertures for 12 cities were taken to be representative of temperatures for the regions around them. The predictand series then consisted of annual January values for each representative city. A separate regression equation was derived for the annual departures from the long term means for January for each city. The procedure therefore predicts the inter-annual variability for January. The regression equations for inter-annual variability were able to explain a much higher percentage of variance compared to the corresponding regression equations for monthly variability. The same procedure can be applied to obtain prediction equations for other months. The higher percentage is to be expected if it is considered that there must be differences in the operation of temperature-controlling physical mechanisms which are dependent on the march of the seasons. (ERA citation 06:031185)